Research On Key Issues Of Mobile Content Distribution

Content distribution is the most fundamental function of the Internet.By distributing digital contents from one node to one or more other nodes,it can implement intercommu-nication between network nodes.With the development of wireless technology,more and more mobile devices are also involved in the content distribution,and thus mobile content distribution is developing gradually.However,as the richer node types,the increasing network traffic and the wider application scenes,the design of mobile content distribution is in dire straits to satisfy the demands of users in many aspects,such as the demand for the speed of content distribution in terms of basic performance and the users ' privacy protection in terms of high-level applications.Motivated by the above mentioned problems,this dissertation investigates a rep-resentative large-scale mobile content distribution system.On the basis of in-depth measurement and analysis,we design and implement a scientific,reasonable and effective content distribution strategy,so as to meet the users ' needs for the basic performance of mobile content distribution.Furthermore,this dissertation researches a secure and efficient mobile content distribution method under the specific application scenario,so as to meet the users ' needs for the advanced requirements of mobile content distribution.The research methodology pays special attention to practicality and scientificity,that is to say,we discover practical problems in real systems and scenarios,and the proposed solutions not only have theoretical basis,but also have practical feasibility.Specifically,the major contributions in this dissertation can be summarized as fol-lows:·Measurement discoveries based on the large-scale dataset of mobile content distribution.By analyzing the log records of 1.36 billion mobile content distribution tasks from Xuanfeng download component,we reveal the manifold performance anomalies and their root causes of the mainstream mobile content distribution system.Especially,we find that the diversity of participation time of data sources is the important metric to indicate the performance of mobile content distribution,and(for the first time)quantity the sweet spots among the file size,the number of data sources used and the speed of content distribution.Our work can provide solid experiences and helpful heuristics to the designers of relevant systems.·Mobile content distribution redirector based on machine learning.In order to avoid the performance degradation of mobile content distribution after accelerating,we design a mobile content distribution redirector in client.In our redirector,the decision tree classification algorithm is used to assist the content distribution strategy selection.Then,the feature-based data source selection model is established based on the multi-armed bandit problem in the reinforcement learning to maximize the rewards achieved by adding data sources.Besides,the real-time monitoring and optimization module is performed based on the diversity of participation time of data sources to optimize the performance.·Secure and efficient mobile location trajectory data distribution based on curve con-struction.In order to ensure the security and effectiveness of the mobile location trajectory data distribution in the map service,we set up a rule for uploading location trajectory data in client to reduce the risk of exposing the individual user's moving trajectory,and use the ?-sample in the curve reconstruction to quantify the degree of privacy leakage.Meanwhile,we design the crust-based map reconstruction algorithm which is robust to noise data to provide high-quality map update service for the server side.